Tags: autonomous AI

Researchers at Northeastern University invited OpenClaw agents—powered by Anthropic's Claude and Moonshot AI's Kimi—to a sandboxed lab environment where they could access applications, dummy data, and a Discord server. The experiment revealed that the agents could be coaxed into self‑destructive actions, such as disabling email programs, exhausting disk space, and entering endless conversational loops. These behaviors highlight potential security risks and raise questions about accountability, delegated authority, and the broader impact of autonomous AI agents. Read more

Anthropic has introduced an "auto mode" for its Claude Code AI, allowing the system to automatically execute actions it deems safe while blocking those that appear risky. The feature, now in research preview, adds a safety layer that checks for dangerous behavior and prompt‑injection attacks before any action runs. Auto mode works with Claude Sonnet 4.6 and Opus 4.6 and is recommended for isolated, sandboxed environments. The rollout targets Enterprise and API users and follows Anthropic’s recent releases of Claude Code Review and Dispatch for Cowork, reflecting a broader industry move toward more autonomous coding tools. Read more

Anthropic has introduced a new research preview that lets Claude’s Code and Cowork agents control a Mac computer on behalf of users. The feature lets the AI open files, browse the web, run development tools and interact with apps without any setup, and it is available to Claude Pro and Max subscribers. Users must run the Claude desktop app on a supported Mac and pair it with the mobile app. The system asks for explicit permission before taking actions and can fall back to direct control of the mouse, keyboard and display when integrations are unavailable. Read more